Skip to content

MLView 0.4.0

Latest

Choose a tag to compare

@realmyang realmyang released this 30 Oct 12:28

MLView is a skill for GitHub Copilot, Codex and Claude Code, a document
format, and a VS Code viewer. Your assistant reads an ML project and writes a
diagram of its workflow: steps, connections and findings, with quotes from the
files it read. The viewer opens that diagram in VS Code, linked to the source.

Preview. The format and the viewer can still change between releases, and
fixes go into the next release.

0.4.0 is the first release. Earlier version numbers were development builds;
none was published as a GitHub release, on the VS Code Marketplace or on Open
VSX. The Claude Code plugin could be installed from this repository during
development.

Files

File What it is
mlview-0.4.0.vsix The VS Code extension
mlview-skill-shared.zip The skill as a ZIP, to extract at the root of your project
mlview-skill-claude-code.zip The skill as a ZIP, to extract at the root of your project
SHA256SUMS The SHA-256 of the three files above

The skill, per assistant
says which ZIP is for which assistant.

Install

  • Download the .vsix file. Install it with VS Code's
    Extensions: Install from VSIX... command.
  • Download the skill ZIP for your assistant and extract it at the root of
    your project.
  • For Claude Code, the skill is also packaged as a
    plugin.

To work from a clone of this repository instead, the
README
says how to build the extension and copy the skill into your project.

Check the downloads

Put the three files and SHA256SUMS in one folder and run this there, on
macOS or Linux:

shasum -a 256 -c SHA256SUMS

Each of the three lines it prints should end in OK. On Linux,
sha256sum -c SHA256SUMS does the same.

On Windows, Get-FileHash -Algorithm SHA256 <file> in PowerShell prints the
value of one file; compare it with that file's line in SHA256SUMS.

What is new

Added

  • The mlview skill for GitHub Copilot, Codex and Claude Code. Your assistant
    reads an ML project and writes a diagram of its workflow: steps, connections
    and findings, with quotes from the files it read.
  • The skill's helper checks a draft's structure, cited paths and exact quotes,
    records file hashes, and publishes a *.mlview.json file.
  • The viewer, a VS Code extension that opens a *.mlview.json file as a
    diagram.
  • In the viewer, Enter opens the cited lines beside the diagram.
  • A review walk that goes through a diagram claim by claim, a phase overview,
    and a search over steps, findings, IDs and cited text.
  • In the review walk, a mark on the line of a connection that has an end out
    of view, and the key o, which brings that end into view.
  • Reveal in Diagram, a command for a file that an open diagram cites.
  • A banner on the diagram when a cited file changed after the revision was
    published.
  • A prompt for your assistant that the viewer copies when you ask for a
    refinement, and a list of the changes since the previous revision.
  • Export of the diagram as an SVG or a PNG image.
  • A Claude Code plugin that holds the skill.
  • An installer that copies the skill into a project. Its --doctor option
    inspects the skill locations of a project without writing.

What the checks cover, and what they do not

  • The helper and the extension check a document's structure, that every cited
    path is inside the workspace, the file hashes and the exact quotes.
  • A matching quote shows that the cited lines exist and have not changed. It
    does not show that they support the claim.
  • No automated check calls a model or opens VS Code.
    What is checked, and what is not
    says what the checks cover, and holds the maintainer's dated record of what
    was checked by hand for this version.
  • Treat a diagram as the assistant's reading of the code, and check it against
    the cited lines.